1
$\begingroup$

I'm trying to solve an advection - convection problem using an implicit upwind scheme - you can see here the finite difference discretization used.

Advection convection implicit upwind

I start the model (built from scratch on Scilab) and I run some convergence tests.
The big problem is that the velocity v is enormous - about 200m/s.
The spatial discretization I use is of 2m, and after some tests it works just fine.
However, the time discretization is what really bugs me. I try to run some scripts by discretizing the time as

time = 0:dt:tend

And I do some tests for different timescales, all fractions of tend. More specifically, the array of different timesteps I use is

dtlist = tend/10^(0:5)

So, for example, for tend=1s I'll test for dtlist = [0.00001, 0.0001, 0.001, 0.01, 0.1, 1]s.
I run the test, and what I get is this:
convergence 0.1s
and
convergence 1s
(DT is the temperature difference between the fluid between t=0 and t=tend)
Now... How is it possible that I have such different results, for the same timesteps of 10^-5s and 10^-4s?

In general, is there a general rule to determine the convergence of an implicit upwind problem? I know that the CFL condition doesn't count here, since the scheme is implicit, but maybe there are other conditions I didn't analize...

Thanks everybody!

$\endgroup$
  • $\begingroup$ did you mean advection-"diffusion" in your question ? $\endgroup$ – SAAD Mar 19 '15 at 10:02
  • 1
    $\begingroup$ What is the correct value for DT? (You should get that if you let both the time step and the mesh size go to zero.) $\endgroup$ – Wolfgang Bangerth Mar 20 '15 at 18:02
1
$\begingroup$

You can take a look at the truncation errors in your discretization. For example, for your time discretization, the truncation error would be $$ \frac{\Delta t}{2} \frac{\partial^2 T}{\partial t^2} $$ An implicit scheme guarantees stability but not accuracy, as the truncation errors can be quite high if large time-steps are taken. Moreover, these errors can add up as you run for longer times. The actual acceptable time-step usually depends on the problem at hand.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.